Codex Major Update: 62 Apps, 110 Skills for Office Workflows
OpenAI shipped a major Codex refresh aimed at office work: six role plugins bundling 62 apps and 110 automation skills across data, creative, sales, product design, public equities, and investment banking. Sites turns deliverables into shareable interactive web pages; Annotations lets you select a region and edit only that part. Codex is no longer positioning itself as “just coding”—it wants to run whole workflows.
Keywords: chatgpt, GPT-5.6, gpt tutorial.
Published: June 3, 2026
By the numbers: 5M weekly users, faster growth outside engineering
Codex passed 5 million weekly active users—about 6× growth since the desktop launch in February. Structure matters: non-developers (analysts, marketing, ops, designers, researchers, investors) are roughly 20% of users and growing about 3× faster than developers.
| Work type | Recent week-over-week (approx.) |
|---|---|
| Data analysis | +110% |
| Research | +37% |
| Knowledge outputs (reports, memos, contracts, decks, sheets) | +36% |
60%+ of users now run multiple Codex tasks in a single day (under half in mid-April). That points to penetration in white-collar work, not only “another assistant for programmers.”
Six plugins: job-specific toolchains
Each plugin packages the software a role already lives in. You give a plain-language instruction; Codex orchestrates cross-app steps in the background.
| Plugin | Typical connectors | Example ask |
|---|---|---|
| Data analysis | Snowflake, Tableau, etc. | “Which region’s revenue fell last quarter and why?”—query, analyze, chart |
| Creative | Figma, Canva, etc. | Turn a creative brief into ad variants, product scenes, commerce images |
| Sales | Salesforce, HubSpot, etc. | Prioritize accounts, prep meetings, follow-ups, CRM updates, close plans |
| Product design | Figma, Canva, etc. | Clickable prototypes from wireframes; audit flows from a live URL |
| Public equities | FactSet, S&P, PitchBook, etc. | Earnings work, comps, thesis checks |
| Investment banking | Research & diligence | Client-ready pitches, trading/comps analyses |
More plugins (corp finance, PE, marketing strategy, consulting, legal) are planned, with a partner ecosystem. OpenAI is acting as a cross-SaaS orchestration layer—not replacing one app, but sitting above them.
Sites: one prompt, a shareable interactive page
Sites is in preview for Business and Enterprise. Instead of only handing you a file, Codex can ship a web page your team opens and uses.
Common cases:
- Finance turns an Excel model into a scenario planner leadership can tune in the browser.
- Launch hubs that centralize copy, milestones, owners, and decisions.
- Pre-review pages with product updates, open issues, usage trends, and next steps.
The gap from “I need an interactive app” to “here is the URL” shrinks from weeks of build to minutes of conversation—and pages can be updated as facts change.
Annotations: select, instruct, patch locally
Annotations moved beyond code to documents, spreadsheets, and slides. Highlight the part you want changed; Codex edits only that region.
Examples: verify a data citation in a memo, fix a chart label on a deck, restyle a nav bar on a Sites page. It targets the “second draft” problem without rewriting everything.
Codex and ChatGPT: converging
The bundle signals Codex for everyone who works, not only engineers. Internally, Codex and ChatGPT are on one trajectory—chat as the front door, workflow execution as the product. The industry shift is from Q&A to “say what you need done.”
Who feels pressure first
Vendors with deep data, compliance, and integrations (e.g. Salesforce, Snowflake) keep short-term moats. The squeeze hits middle layers sold on “great UI”: lightweight BI, PM shells, report builders. When users describe outcomes instead of learning a tool for three days, interface polish is a weaker defense. Moats cluster at data/compliance sources or the agent layer that routes tools.
GPT tutorial: master ChatGPT while Codex runs workflows
Codex connects apps; ChatGPT still carries daily drafting, Q&A, and brainstorming. Before GPT-5.6 is everywhere, practice on GPT-5.4 / GPT-5.5:
- Bound the task—sources, output format, and “do not change” sections (same mindset as Annotations).
- Outline long docs first—headings plus one-line goals per section, then expand.
- Split office work into steps—systems to touch and artifacts to produce—so plugin migration is easier later.
- Keep a fixed eval set—same prompts across reasoning tiers; log quality and latency for comparisons when new models ship.
Use the button below to open LimaxAI chat (gpt-5.4 today; pick newer models when they appear).
Summary
| Item | Takeaway |
|---|---|
| Plugins | Six roles; 62 apps + 110 skills |
| Sites | Interactive pages from natural language |
| Annotations | Local edits on docs, sheets, slides |
| Users | 5M WAU; non-dev share and growth rising |
| Direction | Codex + ChatGPT converging on office agents |
| Next step | Clarify tasks in ChatGPT; try GPT on LimaxAI |
The race is shifting from “can it code?” to “can it finish the workflow?” Until then, getting more from the ChatGPT you already have beats chasing every feature name.